# QuoteDesk

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Tseliso Mosiuoa](https://cerebralvalley.ai/u/bushboy)
- **GitHub:** https://github.com/bushboy/quotedesk
- **Demo video:** https://youtu.be/TvJN9EgG6I8
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1135

QuoteDesk: instant WhatsApp quotes for trade businesses, even when the owner is away

Small trade businesses lose leads every day. A customer sends a WhatsApp photo asking “how much for a carport?”, the owner is on site or it’s a Sunday, and by Monday the customer has gone elsewhere.

QuoteDesk is a governed AI agent that answers on the business’s WhatsApp number. It doesn’t send a generic “we’ll get back to you”. It leads the conversation to one of three outcomes:

✅ Quote: when the job is within the owner’s policy, it sends a fixed quote or a price range, worked out from the business’s own rate card
🔍 Book an inspection: when a job needs a site visit, it says why and books one
📨 Escalate: anything else goes to the owner, with the full context so they can pick it up immediately

How it works

The customer can send text, photos or video, in any order, with no menus.
The agent reads the photo and states what it sees, for example “Looks like a double driveway, about 6 × 6 m. Is that right?”. It asks the customer to confirm instead of assuming they know materials or sizes.
Prices, dates and conditions come only from deterministic policy code, never from the model. The model reads the input and writes the reply; the code makes every commitment.
Owners set their rate card and rules in a simple console, and approve exceptions there.
It’s tested against scripted customer conversations for each business before it goes near a live number.

Built with Meta models

Muse Spark 1.3: conversation, fact extraction and image understanding
SAM 3.1: segmenting the object in the customer’s photo as evidence for the quote
Muse Image 1.0: Generating artist impressions

Built in Python (FastAPI, Postgres with row-level security) for the Meta Global AI Developer Hackathon, Indie track.

The demo businesses, FrameWorks (carports) and AquaSure (water tanks), are fictional.  The FrameWorks Demo is the most advanced of the two.  The judges should only run that one.

#MetaAI #WhatsApp #AIagents #SmallBusiness #SouthAfrica

## More from Meta Global AI Developer Hackathon

- [ROOKWELD](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1132)
- [PuzzleAR](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1133)
- [Iqbal's Basilisk Lab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1134)
- [Nanang](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1136)
- [KineticLab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1137)
- [Esmada Zimba](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1138)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1135. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt